How accurately can sitting and the intensity of walking and cycling be classified using an accelerometer on the waist for the purpose of the “Global recommendations on physical activity for health”?

The society has become more sedentary and has developed a lack of physical activity, therefore increasing health risks. Feedback is needed to change these behaviours. For this feedback, first accurate monitoring is needed: sedentary behaviour must be classified as well as the intensity of physical activity. In this report a State of the Art analysis is performed to compare different classification techniques and finally two methods, both using an accelerometer on the waist, are worked out. These methods are Integral of the Modulus of the Accelerometer (IMA) classification and a machine learning technique (MLT): support vector machine (SVM). These methods are then applied in a laboratory experiment to study their quality. A measurement setup is made to create a dataset of the following activities: standing, sitting, lying, walking (2.4 - 7.5 km/h) and cycling (10.1-19 km/h). This dataset (n=15) is analysed and classified using Matlab for both methods. The IMA method was unable to monitor sedentary behaviour, but could classify the physical activity (PA) intensity with an accuracy of 66%. The SVM method within subjects was able to monitor sedentary behaviour with an accuracy of 91±20% and the classification of the PA intensity has an accuracy of 94±5%. For between subjects the accuracies decrease to 71±13% for PA intensity accuracy and 45±35% for the sedentary behaviour classification. IMA was implemented in the old feedback system, monitoring the overall daily amount of physical activity, but can significantly be outperformed by replacing it with the current SVM implementation. At this moment however, SVM can only be used to improve the old system, it cannot yet be used to create new additions to the feedback system, such as the implementation of the feedback of the sedentary behaviour and specific physical activity intensities. New methods and properties of the current SVM system have been found that might increase the accuracies of the between subject analysis and therefore might enable SVM to become applicable for the extra feedback options.

[1]  Ifeyinwa E. Achumba,et al.  Real-Time Human Ambulation, Activity, and Physiological Monitoring: Taxonomy of Issues, Techniques, Applications, Challenges and Limitations , 2013, Sensors.

[2]  Shane A Lowe,et al.  Monitoring human health behaviour in one's living environment: a technological review. , 2014, Medical engineering & physics.

[3]  Rae-Woong Park,et al.  Different Location of Triaxial Accelerometer and Different Energy Expenditures , 2014, Yonsei medical journal.

[4]  B E Ainsworth,et al.  Compendium of physical activities: an update of activity codes and MET intensities. , 2000, Medicine and science in sports and exercise.

[5]  Pierre Jallon,et al.  Automatic identification of physical activity types and sedentary behaviors from triaxial accelerometer: laboratory-based calibrations are not enough. , 2015, Journal of applied physiology.

[6]  Adil Mehmood Khan,et al.  Activity Recognition on Smartphones via Sensor-Fusion and KDA-Based SVMs , 2014, Int. J. Distributed Sens. Networks.

[7]  Davide Anguita,et al.  Human Activity Recognition on Smartphones Using a Multiclass Hardware-Friendly Support Vector Machine , 2012, IWAAL.

[8]  S. Intille,et al.  Estimating activity and sedentary behavior from an accelerometer on the hip or wrist. , 2013, Medicine and science in sports and exercise.

[9]  D. Bassett,et al.  The technology of accelerometry-based activity monitors: current and future. , 2005, Medicine and science in sports and exercise.

[10]  Christos Pilagkas.  Activity classification using a hip worn inertial sensor , 2012 .

[11]  Gert R. G. Lanckriet,et al.  A random forest classifier for the prediction of energy expenditure and type of physical activity from wrist and hip accelerometers , 2014, Physiological measurement.

[12]  Kenneth Meijer,et al.  Activity identification using body-mounted sensors—a review of classification techniques , 2009, Physiological measurement.